The Design And Implementation Of Fairshare Scheduling Module In Grid Platform

In the grid environment, computing resources, storage resources, data resources can be fully shared. In a shared environment, it happens very often that vicious users compete for limited resources. It can easily lead to an imbalance that low-priority users or low-priority scheduling queue have no chance to use those resources.This thesis is based on the existing grid computing platform product. By modifying and expanding the former scheduling policy, the thesis enhances the product with fairshare scheduling strategy which user can select and configure. This can help customers to solve the problems that low-priority users have a very low usage of the resources and low-priority job queues have very low throughput in the enterprise grid environment. It also can save the system administrator's time and energe to manage cluster and user. Balanced usage of resources can improve enterprises productivity.This paper follows the existing system architecture and scheduling module framework of the grid computing product, developed two strategies: the user-based and queue-based fairshare scheduling strategy. Meanwhile we develop three user-based scheduling sub-strategies including host level, the queue level and cross-queue level. To enhance system fault tolerance and maintainability, we introduce the log management for scheduling module. Among them, the fairshare scheduling strategy is developed by plugins and integrated into existing scheduling module by modifying the logic of related modules in the grid computing platform product; data required in fairshare scheduling strategy, configuration and other information are collected and realized by expanding the original system function module; log management system is realized through using bit computing which is new and different with the original system ways.By the corporation of scheduling policy, other various auxiliary functions and development, the user-based fairshare scheduling strategy reduces the time which low-priority users wait for grid resources, balancing competition for resources between the high-priority users and low-priority users. Among them, host level sub-strategy balances the opportunity to use computers between the different priorities users, shortening the wait time for low priority users to use the system; queue level and cross-queues sub-strategies balance the different priorities users on the queue to compete the opportunity dispatching theirs jobs. Through the implementation, it reduces the high-priority user's priority and improves the low-priority user's priority according to their resources utilization rate. Queue-based fairshare scheduling strategy, through implementing, reduces the time which the low-priority queues wait for dispatching their jobs and improves the low-priority queue's throughput of dispatching jobs. It balances competition for resources between the high priority and low priority queues in the network grid environment. The log management allows users to see different levels of system operation information, in order to track system status and diagnose system exceptions.